Modern industrial system has been developed into the direction of more and more larger and complex, which makes the fault diagnosis for industrial system present a series of technical problems. In recent years, deep learning has shown its unique potentials and advantages in feature extraction and pattern recognition. And the application of deep learning to achieve fault diagnosis of complex industrial systems has begun on its initial exploration stage. In this paper, several typical methods based on deep learning have been introduced first, which can be employed to realize the fault diagnosis for industrial system. And then, the main idea and model approach for fault diagnosis based on deep learning have been described. Finally, the characteristics of faults in complex industrial system have been summarized and discussed, and the challenge of deep learning in realizing the fault diagnosis for complex industrial system has also been studied and discussed.